Advancing Colorimetric Analysis in Enzyme-Linked Immunosorbent Assays: Harnessing Nonlinear Regression for Improved...

Smartphone-based colorimetric enzyme-linked immunosorbent assay (ELISA) readers have emerged as a cost-effective and portable alternative to conventional spectrophotometric systems, especially for use in resource-limited and point-of-care settings. In our previous work, we developed a digital image colorimetry platform integrated with a 3D-printed optomechanical system and smartphone imaging, achieving high diagnostic accuracy for cancer cell lines through linear regression modeling of red, green, and blue (RGB) intensities. While effective, this linear approach could not fully capture the nonlinear relationships intrinsic to enzymatic colorimetric reactions and light-matter interactions. In this study, we present a significant enhancement to our prior model by implementing a nonlinear machine learning framework, eXtreme gradient boosting (XGBoost), to better predict optical densities from smartphone-captured RGB data. Utilizing the same hardware platform and experimental protocol as our earlier system, we extracted RGB values from ELISA images under RGB backlighting and then expanded the feature space through engineered transformations. These features were used to train and validate the XGBoost model on multiple human cancer cell lines (HE4, PC3, 5637, and ACHN). Our XGBoost model achieved exceptional predictive performance with R ² > 0.999 across all tested cell lines, substantially improving upon the linear regression model (R ² range: 0.923-0.996). Root mean square error was reduced by over 98%, and Spearman and Pearson correlation coefficients approached unity, demonstrating excellent trend fidelity and linear alignment with FDA-certified ELISA readers (Epoch and Tecan). The model's robustness was further confirmed through k-fold cross-validation and rigorous statistical analysis. This advancement positions our platform as a powerful candidate for decentralized diagnostics and high-throughput screening, capable of operating reliably across different smartphones and lighting conditions.